Skip to content
Montex Alpha Rail System
$497

Monte Carlo Lab · The R&D Boundary

A sandbox with a paper trail.

Runs logged

ALL

Original workbook

UNTOUCHED

Gate before run

READY

How MARS uses this

The R&D boundary exists because simulated territory includes places the live account must never casually visit. The depth-duration map makes the boundary concrete: experiments may explore the deep-slow tail freely, but nothing graduates to live authority while its distribution still lives there.

How it benefits you

Drawdown stops being one scary number. A 7% dip reads as the common, quick-recovery event the map says it is, while the non-linear duration curve makes the case for gate compression viscerally: deep holes do not just cost more - they cost longer.

8w16w24w32wrecovery time grows non-linearlydepth →−3%−9%−16%−23%−28%bubble size = how often simulated paths visit that territory

The deep-slow tail is legal territory for research and forbidden territory for live rules — the boundary in one picture.

Seed discipline

The seed is what turns two runs into a comparison.

Changing a structure and re-running with a fresh random draw produces two numbers whose difference is part structure and part luck, in unknown proportion. Holding the seed fixed removes the luck term: the same path sequence meets both structures, so the delta is attributable. This is the difference between testing a change and observing one. Command-line flags can override the workbook values for paths, seed, and detail, and the override is written into the run note so the log never disagrees with what actually ran.

The gate

BLOCKED names the zone. It does not ask for a judgement call.

The Validation tab is the single green light for the workbook, aggregating the probability integrity checks from Zone 3, the expected-value GO or NO-GO from Zone 4, and the calibration integrity checks from Zone 7. When it reads BLOCKED, the flagged row names the zone at fault rather than describing a mood. The operator fixes the flagged zone, recalculates, and confirms READY. No run should proceed while the gate is red, because a run built on a broken probability ladder produces output that is internally consistent and externally meaningless.

  • Zone 3 checks monotonicity, range, blank safety, and that branch weights sum to 1.00.
  • Zone 4 computes per-branch and blended expectancy and returns a GO or NO-GO verdict before any path is simulated.
  • Zone 7 flags missing data, invalid labels, and insufficient evidence in the calibration bridges.

Confidence before conclusion

A marginal result gets its error bands read before it gets believed.

Table AF reports path stability, a confidence grade, and Monte Carlo error bands for the run. The reading order after a marginal outcome is fixed: read the executive interpretation, read the question engine, look at the shape in the equity and drawdown fans, drill into branch and variant attribution, and then, before acting, check that the result is statistically trustworthy at the path count used. A conclusion drawn from an unstable run is not a finding. It is a sampling artifact wearing a table's formatting.

What crosses the boundary

04

Simulation informs sizing. It never authorises it.

Nothing produced in the Lab expands live risk. The benchmark standard states the boundary plainly: the model does not override gate or brake authority, and drawdown state remains the highest capital authority. A scenario that survives beautifully in resampling still enters live rules only through the promotion workflow, and even a promoted change operates inside whatever gate the live account is currently in.

Further illustration

Four scenario profiles judged side by side: expectancy, adverse-tail cost, and a verdict. Changes graduate through this grid or not at all.

BASELINE

LIVE
EV +0.31RP10 DD −14.8%

current governing profile

TIGHTER STOPS

REJECTED
EV +0.24RP10 DD −11.2%

EV cost exceeds drawdown saving

TNP WEIGHT +10

SANDBOX
EV +0.36RP10 DD −19.6%

edge up, adverse tail deepens

FEE MODEL B

CANDIDATE
EV +0.33RP10 DD −14.9%

friction saving survives resampling

Reference

The run contract: what each artifact guarantees

ArtifactWhere it livesWhat it guarantees
Input workbookMARS_MC_Lab_v2_5.xlsxNever written to by the engine, so configuration cannot drift inside a run
Output workbook..._RUN.xlsxResults are values, not formulas, and belong to exactly one execution
Random seedParameter ConsoleA fixed seed makes a run reproducible and makes two structures comparable like for like
Engine run logEngine_Run_LogOne appended row per execution with ID, seed, scenario, and headline results
Validation gateZone 12 ValidationAggregates every zone's own checks into a single READY or BLOCKED verdict
Run confidenceTable_AF_RunConfidencePath stability, confidence grade, and Monte Carlo error bands for the result
Archive notesArchive_NotesDeprecated assumptions and prior parameter sets, so model history survives

Edge cases & failure modes

Where it can mislead

  • !A run with an unlogged manual flag override is not evidence, because the log and the run no longer agree.
  • !Comparing two structures under different seeds measures structure plus luck, and cannot separate them.
  • !A green result at low path count can still carry error bands wide enough to invert the conclusion.

Doctrine

Connected inside MARS

This module doesn't work alone.

Connected — 01

Monte Carlo Lab (full module)

Open system page →

Connected — 02

Scenario Manager

Open system page →

Connected — 03

Throttle Control Panel

Open system page →

Go deeper

Operator briefs on this territory.

Deep dive — 01

The research wall, implemented as an immutability guarantee.

The engine cannot write to the source workbook. The wall is a file permission, not a promise.

Read the full brief →

Deep dive — 02

Reproducibility is what turns a run into evidence.

Same seed, different structure: the only honest way to ask whether a change did anything.

Read the full brief →

Deep dive — 03

Ask whether the run is trustworthy before asking what it says.

Path count buys precision, not truth. When the margin is thin, the error band decides.

Read the full brief →

Take it further

The research boundary exists because production rules must not move on a hunch. The Foundry is where a hunch can be taken seriously without ever touching a live rule.

Open the AlphaRail Foundry Lab ↗

Every module ships in the complete MARS package.

One price. Eleven workbooks, three TradingView indicators, and the full manual library — $497.

Get MARS Pro — $497Explore the Platform